search

wjun0830 / QD-DETR

Official pytorch repository for "QD-DETR : Query-Dependent Video Representation for Moment Retrieval and Highlight Detection" (CVPR 2023 Paper)
https://arxiv.org/abs/2303.13874
Other
199 stars 15 forks source link

TVSUM data issue #32

Closed yaokunZ closed 1 year ago

yaokunZ commented 1 year ago

image Why do we need to divide by 80 and multiply by 12?

wjun0830 commented 1 year ago

We did that to normalize the saliency scores to a scale similar to the QVhighlights saliency score.

yaokunZ commented 1 year ago

But I found that the video length is not 80, why does normalization need to be divided by 80? image

wjun0830 commented 1 year ago

It's not the length. We normalized the scale of the saliency score

2023년 9월 20일 (수) 오후 2:42, zyk @.***>님이 작성:

But I found that the video length is not 80, why does normalization need to be divided by 80? [image: image] https://user-images.githubusercontent.com/55798047/269173009-9dbe4d75-417c-46f1-ad70-07e7f0762ae9.png

— Reply to this email directly, view it on GitHub https://github.com/wjun0830/QD-DETR/issues/32#issuecomment-1727006483, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHQYPMHDH4FJHOBHSUH6MDDX3J645ANCNFSM6AAAAAA46RQVYE . You are receiving this because you commented.Message ID: @.***>

yaokunZ commented 1 year ago

Thanks! There are 20 scores per clip in TVSUM , but QVhighlights only has 3 saliency scores.